The Replica Consistency Problem in Data Grids

Fast and reliable data access is a crucial aspect in distributed computing and is often achieved using data replication techniques. In Grid architectures, data are replicated in many nodes of the Grid, and users usually access the "best" replica in terms of availability and network latency. When replicas are modifiable, a change made to one replica will break the consistency with the other replicas that, at that point, become stale. Replica synchronisation protocols exist and are applied in several distributed architectures, for example in distributed databases. Grid middleware solutions provide well established support for replicating data. Nevertheless, replicas are still considered read-only, and no support is provided to the user for updating a replica while maintaining the consistency with the other replicas. In this thesis, done in collaboration with the Italian National Institute of Nuclear Physics (INFN) and the European Organisation for Nuclear Research (CERN), we study the replica consistency problem in Grid computing and propose a service, called CONStanza, that is able to synchronise both files and heterogeneous (different vendors) databases in a Grid environment. We analyse and implement a specific use case that arises in high energy Physics, where conditions databases are replicated using databases of different makes. We provide detailed performance results, and show how CONStanza can be used together with Oracle Streams to provide multitier replication of conditions databases using Oracle and MySQL databases.